U.S. patent application number 11/810704 was filed with the patent office on 2008-12-11 for segmenting colon wall via level set techniques.
This patent application is currently assigned to The Government of the U.S.A. as represented by the Secretary of the Dept. of Health & Human Services. Invention is credited to Ingmar Bitter, Ronald M. Summers, Robert L. Van Uitert, JR..
Application Number | 20080304616 11/810704 |
Document ID | / |
Family ID | 40095878 |
Filed Date | 2008-12-11 |
United States Patent
Application |
20080304616 |
Kind Code |
A1 |
Van Uitert, JR.; Robert L. ;
et al. |
December 11, 2008 |
Segmenting colon wall via level set techniques
Abstract
Various level set techniques can be used to automatically
segment the colon wall, including identifying the colon wall outer
boundary. A speed image can be used during level set processing.
For example, the speed image can be generated via inverting the
gradient perpendicular to the segmented inner boundary of the colon
wall. The techniques can be useful for determining wall thickness,
which can be used to classify polyp candidates, diagnose diseases
of the colon, and the like.
Inventors: |
Van Uitert, JR.; Robert L.;
(Germantown, MD) ; Summers; Ronald M.; (Potomac,
MD) ; Bitter; Ingmar; (Maple, CA) |
Correspondence
Address: |
KLARQUIST SPARKMAN, LLP
121 S.W. SALMON STREET, SUITE #1600
PORTLAND
OR
97204-2988
US
|
Assignee: |
The Government of the U.S.A. as
represented by the Secretary of the Dept. of Health & Human
Services
|
Family ID: |
40095878 |
Appl. No.: |
11/810704 |
Filed: |
June 5, 2007 |
Current U.S.
Class: |
378/4 ; 378/8;
382/131; 600/407 |
Current CPC
Class: |
A61B 6/03 20130101; G06T
2207/20161 20130101; G06T 7/149 20170101; G06T 2207/30028 20130101;
G06T 7/12 20170101; G06T 2207/10081 20130101; G06T 7/0012
20130101 |
Class at
Publication: |
378/4 ; 378/8;
600/407; 382/131 |
International
Class: |
A61B 6/03 20060101
A61B006/03 |
Claims
1. A computer-implemented method comprising: receiving a digital
representation for a colon, wherein the digital representation
represents at least a portion of a colon wall for the colon;
identifying an outer boundary for the colon wall via a level set
technique; and outputting an indication of the outer boundary of
the colon wall.
2. One or more computer-readable media comprising
computer-executable instructions causing a computer to perform the
method of claim 1.
3. The computer-implemented method of claim 1 wherein the
indication of the outer boundary of the colon wall is of subvoxel
accuracy.
4. The computer-implemented method of claim 1 wherein the level set
technique comprises three-dimensional geodesic active contour level
set segmentation.
5. The computer-implemented method of claim 1 wherein: identifying
the outer boundary for the colon wall via the level set technique
comprises: using a lumen segmentation of the digital representation
as an initial level set boundary.
6. The computer-implemented method of claim 1 further comprising:
identifying an inner boundary of the colon wall; and via the inner
boundary of the colon wall and the outer boundary of the colon
wall, calculating a thickness of the colon wall.
7. The computer-implemented method of claim 6 wherein: the outer
boundary of the colon wall is represented as a surface; the inner
boundary of the colon wall is represented as a surface; and
calculating thickness comprises determining distance between the
surfaces.
8. The computer-implemented method of claim 6 further comprising:
submitting a set of characteristics for a polyp candidate to a
polyp candidate classifier configured to determine whether the
polyp candidate is a true positive; wherein the set of
characteristics comprises the thickness of the colon wall.
9. The computer-implemented method of claim 1 further comprising:
segmenting the colon wall.
10. The computer-implemented method of claim 9 further comprising:
determining a thickness of the colon wall.
11. The computer-implemented method of claim 9 further comprising:
detecting colonic diverticular disease via the thickness of the
colon wall.
12. The computer-implemented method of claim 9 further comprising:
detecting colon spasm via the thickness of the colon wall.
13. The computer-implemented method of claim 9 further comprising:
detecting colon cancer via the thickness of the colon wall.
14. The computer-implemented method of claim 9 further comprising:
detecting presence of a polyp via the thickness of the colon
wall.
15. The computer-implemented method of claim 1 wherein: identifying
the outer boundary for the colon wall comprises identifying at
least a location of the outer boundary of the colon wall; and the
indication of the outer boundary of the colon wall indicates at
least the location of the outer boundary of the colon wall.
16. The computer-implemented method of claim 1 wherein the level
set technique comprises: generating a speed image; and evolving an
isosurface based at least on the speed image.
17. The computer-implemented method of claim 16 wherein generating
the speed image comprises: calculating a directional derivative of
the digital representation in a direction perpendicular to a colon
wall inner boundary represented in a colon wall inner boundary
segmentation.
18. The computer-implemented method of claim 17 further comprising:
suppressing local non-maximum gradients along a level set expansion
direction via a sigmoid filter.
19. The computer-implemented method of claim 17 further comprising:
performing a lumen segmentation for the digital representation; and
using a lumen boundary in the lumen segmentation as the colon wall
inner boundary segmentation.
20. The computer-implemented method of claim 19 wherein the lumen
segmentation comprises: generating a threshold region growing
segmentation; and segmenting via a threshold level set technique,
wherein the threshold level set technique uses the threshold region
growing segmentation as an initial level set boundary.
21. The computer-implemented method of claim 16 wherein: the speed
image is determined via the digital representation for the
colon.
22. The computer-implemented method of claim 1 further comprising:
segmenting an entire colon wall of the colon.
23. An apparatus comprising: means for receiving a digital
representation for a colon, wherein the digital representation
represents at least a portion of a colon wall for the colon; means
for identifying an outer boundary for the colon wall via level set
processing; and means for outputting an indication of the outer
boundary of the colon wall.
24. A computer-implemented method comprising: receiving a digital
representation for a colon, wherein the digital representation
represents at least a portion of a colon wall for the colon;
segmenting the colon wall via a level set technique, whereby the
segmenting results in a segmented colon wall for the digital
representation; and via the segmented colon wall for the digital
representation, automatically calculating colon wall thickness for
at least a portion of the colon wall.
25. A computer-implemented method comprising: receiving a
three-dimensional digital representation for a colon, wherein the
digital representation comprises a computed tomography image
representing at least a portion of a colon wall for the colon; from
the digital representation for the colon, generating a lumen
segmentation indicating a boundary of an inner wall of the colon;
producing lumen segmentation level sets from the lumen
segmentation; from the lumen segmentation and the computed
tomography image, generating a speed image via a three-dimensional
derivative of the computed tomography image in a direction
perpendicular to lumen segmentation level sets, wherein local
non-maximum gradients along level set expansion direction are
suppressed, and generating the speed image comprises applying a
sigmoid filter emphasizing high directional derivatives and
inverting the speed image; generating a level set image via a level
set segmentation of an outer wall of the colon via
three-dimensional geodesic active contour level set segmentation
with the speed image, wherein the lumen segmentation level sets are
used as an initial level set boundary, an advection term attracts
level set evolution to high gradient values, and a curvature term
prevents evolution of the boundary from exceeding a maximum
curvature; determining a boundary of the outer wall of the colon
via an isocontour in the level set image; and outputting an
indication of the boundary of the outer wall of the colon.
26. A computer-implemented method comprising: receiving a digital
representation for an anatomical structure, wherein the digital
representation represents at least a portion of a wall for the
anatomical structure; identifying an outer boundary for the wall
via a level set technique; and outputting an indication of the
outer boundary of the wall.
27. The computer-implemented method of claim 26 further comprising:
determining a thickness of the wall.
28. The computer-implemented method of claim 27 further comprising:
detecting atherosclerosis via the thickness of the wall.
29. The computer-implemented method of claim 27 further comprising:
detecting hyperplasia via the thickness of the wall.
30. A computer-implemented method comprising: receiving a digital
representation for a colon, wherein the digital representation
represents at least a portion of a wall for the colon; determining
wall thickness for the wall; and detecting presence of colonic
diverticular disease via determined wall thickness for the
wall.
31. The method of claim 30 wherein determining wall thickness for
the wall comprises applying a level set technique to the digital
representation.
32. The method of claim 30 wherein determining wall thickness for
the wall comprises applying a binary space partitioning tree.
33. The method of claim 30 wherein detecting presence of colonic
diverticular disease comprises: clustering candidate detections
within a threshold distance.
34. The method of claim 30 wherein detecting presence of colonic
diverticular disease comprises: responsive to determining that a
location on the colon is under a threshold thickness, classifying
the location as not having diverticular disease.
35. The method of claim 30 wherein detecting presence of colonic
diverticular disease comprises: responsive to determining that CT
intensity of a location on an outer wall of the colon is outside of
a normal range for colon wall, classifying the location as not
being a detection of diverticular disease.
36. The method of claim 30 wherein detecting presence of colonic
diverticular disease comprises: computing an average and standard
deviation of colon wall thickness for a cluster of locations.
37. The method of claim 30 wherein detecting presence of colonic
diverticular disease comprises: computing an average and standard
deviation of intensity values for a cluster of locations.
38. The method of claim 30 wherein detecting presence of colonic
diverticular disease comprises: providing a plurality of features
for a cluster of locations to a support-vector machine classifier
configured to indicate whether the cluster of locations are
indicative of colonic diverticular disease.
39. One or more computer-readable media comprising
computer-executable instructions causing a computer to perform the
method of claim 30.
Description
TECHNICAL FIELD
[0001] The field relates to software analysis of images in a
medical context.
BACKGROUND
[0002] Although colon cancer is the second leading cause of cancer
death in the United States, it is also often treatable. Early
detection of colon polyps is a key to treatment. CT colonography
(CTC), also known as virtual colonoscopy, is a promising new
non-intrusive detection technique where polyps are identified from
computed tomography (CT) scans, sometimes with the aid of a
computer-aided detection (CAD) system.
[0003] While the inner boundary of the colon wall has often been
the focus of previous colon segmentation work, detection of the
colon wall outer boundary is often difficult due to the low
contrast between CT attenuation values of the colon wall and the
surrounding fat tissue.
[0004] Thus, more work is needed to better detect the colon wall
outer boundary and otherwise improve virtual colonoscopy
technologies.
SUMMARY
[0005] A digital representation for an anatomical structure can be
processed via a level set technique to identify the colon outer
wall boundary. For example, the colon wall outer boundary can be
segmented.
[0006] A speed image for the level set technique can be generated
from a segmentation of the colon wall inner boundary. For example,
the speed image can be generated via a gradient along a vector
(e.g., perpendicular to the colon wall inner boundary) in a digital
representation for a colon. The speed image can be used during
level set processing to identify the colon wall outer boundary.
[0007] The segmented colon wall outer boundary can be used for a
variety of purposes. For example, colon wall thickness can be
determined. Colon wall thickness can be used in polyp candidate
identification and classification, and diagnosis of colonic
diseases (e.g., detection of diverticular disease). Other uses of
the segmented colon wall include spasm detection, cancer detection,
colon centerline determination, and fly throughs.
[0008] Additional features and advantages of the technologies
described herein will be made apparent from the following detailed
description of illustrated embodiments, which proceeds with
reference to the accompanying drawings.
BRIEF DESCRIPTION OF THE DRAWINGS
[0009] The patent or application file contains at least one drawing
executed in color. Copies of this patent or patent application
publication with color drawings will be provided by the Office upon
request and payment of the necessary fee. FIGS. 16-22, 24-30, and
33-35 are executed in color.
[0010] FIG. 1 is a block diagram of an exemplary system configured
to process a digital representation for a colon and output an
indication of a colon wall outer boundary via a level set
technique.
[0011] FIG. 2 is a flowchart of an exemplary method of processing a
digital representation for a colon and outputting an indication of
a colon outer wall boundary via a level set technique and can be
implemented, for example, in a system such as that shown in FIG.
1.
[0012] FIG. 3 is a block diagram of an exemplary system configured
to process a digital representation for a colon and output an
indication of a colon outer wall boundary via a speed image.
[0013] FIG. 4 is a flowchart of an exemplary method of processing a
digital representation for a colon via a level set technique with a
speed image and outputting an indication of a colon outer wall
boundary.
[0014] FIG. 5 is a block diagram of an exemplary system configured
to process a digital representation for a colon and output an
indication of colon wall thickness therefrom.
[0015] FIG. 6 is a flowchart of an exemplary method of processing a
digital representation for a colon and outputting an indication of
colon wall thickness therefrom.
[0016] FIG. 7 is a block diagram of an exemplary system configured
to process a digital representation for a colon and an inner wall
boundary segmentation to generate a speed image.
[0017] FIG. 8 is a flowchart of an exemplary method of generating a
speed image from a digital representation for a colon and an inner
wall boundary segmentation.
[0018] FIG. 9 is a block diagram of an exemplary system configured
to process a digital representation for a colon and generate an
indication of a colon wall outer boundary.
[0019] FIG. 10 is a flowchart of an exemplary method of generating
an indication of a colon wall boundary via a speed image and a
colon inner wall boundary segmentation.
[0020] FIG. 11 is a two-dimensional CT image slice for a colon
showing colon wall.
[0021] FIG. 12 is a detail of FIG. 11 showing the colon outer wall
and low contrast between the colon outer wall and the surrounding
tissue.
[0022] FIG. 13 is a two-dimensional CT image slice showing
superimposed results of lumen segmentation.
[0023] FIG. 14 is a detail of a two-dimensional CT image slice
showing superimposed level set isocontours from the lumen level set
segmentation.
[0024] FIG. 15 is a graph showing an intensity profile of a CT
image and gradient magnitude along a directed ray from the colon
lumen to the outer wall.
[0025] FIG. 16 shows a representation of a colon with a cutting
plane showing segmentation of the lumen and outer colon wall.
[0026] FIG. 17 is a two-dimensional CT image slice showing
superimposed results of colon wall segmentation.
[0027] FIG. 18 is a two-dimensional CT image slice showing
superimposed results of colon wall segmentation.
[0028] FIG. 19 is a two-dimensional CT image slice showing
superimposed results of colon wall segmentation.
[0029] FIG. 20 is a detail of a two-dimensional CT image slice
showing superimposed results of colon wall segmentation.
[0030] FIG. 21 is an exemplary calculated speed image used during
level set segmentation of the colon outer wall.
[0031] FIG. 22 is a two-dimensional CT image slice showing
superimposed results of colon wall segmentation via the speed image
shown in FIG. 21.
[0032] FIG. 23 is an illustration of a surface of an outer colon
wall determined via level set segmentation.
[0033] FIGS. 24, 25, and 26 are illustrations of colons indicating
thickness of colon wall determined via level set segmentation.
[0034] FIGS. 27A-C are illustrations of colons indicating thickness
of colon wall determined via level set segmentation with regions of
interest therein shown in respective insets.
[0035] FIGS. 28, 29, and 30 are illustrations of colon cross
sections indicating polyp detections based on colon wall thickness
determined via level set segmentation.
[0036] FIG. 31 is a slice of a CTC scan showing colonic wall
affected by diverticular disease.
[0037] FIG. 32 is an inset of FIG. 31.
[0038] FIG. 33 is a slice of a CTC scan showing segmentation of the
colon wall.
[0039] FIG. 34 is a slice of a CTC scan showing the sigmoid portion
of a colon with diverticular disease.
[0040] FIG. 35 is a slice of a CTC scan showing the sigmoid portion
of another colon with diverticular disease.
[0041] FIG. 36 is a graph of a free-response receiver operating
characteristic curve for four features calculated on diverticular
detection candidates.
[0042] FIG. 37 is a block diagram of an exemplary computer system
for implementing the described technologies.
DETAILED DESCRIPTION
Overview of Technologies
[0043] The technologies described herein can be used in any of a
variety of scenarios in which identifying the colon wall is
desired. For example, when performing computer-aided detection of
polyps in a CT scan of the colon, colon wall thickness can be
considered when identifying or classifying candidate polyps.
Further, diseases of the colon can be diagnosed via colon wall
thickness calculations. For example, diverticular disease can be
detected. Still further, the technologies described herein can be
used for colon spasm detection, colon cancer detection, colon
centerline determination, and flythroughs.
[0044] A digital representation for an anatomical structure
includes any digital representation of an anatomical structure (or
portion thereof) stored for processing in a digital computer. For
example, representations can include two- or three-dimensional
representations (e.g., one or more images) of portions of an
anatomical structure stored via a variety of data structures.
Representations can be composed of pixels, voxels, or other
elements. A digital representation of an anatomical structure is
sometimes called "virtual" (e.g., a "virtual colon") because it is
a digital representation that can be analyzed to learn about the
represented anatomical structure.
[0045] A component of a digital representation includes any two- or
three-dimensional element that composes a part of a representation
of a portion of an anatomical structure stored as an image. For
example, pixels and voxels can be components.
[0046] Segmentation includes the process of dividing a digital
representation for an anatomical structure into constituent parts
into which a body, entity, or quantity is divided or marked off by
or as if by natural boundaries. Thus, segmentation can include
identifying the boundaries of an anatomical structure. Segmentation
can include identifying the colon wall outer boundary. Further,
segmentation can determine the location and extent of an anatomical
structure or its boundary. For example, segmentation can indicate
which portions of a digital representation are part of a colon
wall, and which parts are not part of the colon wall. Types of
segmentation include freehand segmentation, region-based (or
region-growing) segmentation, fuzzy connectedness segmentation,
K-means clustering segmentation, level set segmentation, active
contours segmentation, expectation-maximization segmentation, and
so on.
[0047] Imaging includes any technologies for obtaining an image of
the inside of a body by transmitting electromagnetic or sonic waves
through the body. Imaging includes radiographic images (with
X-rays, for example computer tomography or "CT"), sonic energy
(such as ultrasound) and magnetic fields (such as magnetic
resonance imaging, or "MRI"). Although representations of an
anatomical structure using such technology are sometimes called an
"image," in practice, the representation can be a series of image
slices (e.g., two-dimensional image slices stacked together to form
a three-dimensional representation).
[0048] Exemplary anatomical structures in any of the examples
herein include such structures as the colon, heart, bronchi, blood
vessels, small bowel, biliary tract, urinary tract, and
esophagus.
EXAMPLE 1
Exemplary System Outputting an Indication of Colon Wall Outer
Boundary
[0049] FIG. 1 is a block diagram of an exemplary system 100
configured to process a digital representation for a colon and
output an indication of a colon wall outer boundary via a level set
technique. In the example, a digital representation 112 for a colon
is processed by software 122 (e.g., employing a level set
technique, such as any of the level set techniques described
herein), which outputs an indication 182 of a colon wall outer
boundary.
EXAMPLE 2
Exemplary Colon Wall
[0050] In any of the examples herein, the inner boundary of the
colon wall can be a boundary between the colon wall and what is
inside the colon (e.g., the lumen-mucosal boundary). The lumen
boundary (e.g., boundary between the lumen and the colon) can be
used as the colon wall inner boundary, and references to the lumen
can imply the lumen boundary.
[0051] The outer boundary of the colon wall can be a boundary
between the colon wall and what is outside the colon. For example,
the outer boundary can be the colon serosal-tissue boundary,
serosal soft-tissue boundary, serosal-fat boundary, serosal-organ
boundary, serosal-serosal boundary (e.g., for two bowel loops that
abut), or the like.
[0052] For the sake of convenience, sometimes the inner boundary of
the colon wall is called the "inner wall," and the outer boundary
of the colon wall is called the "outer wall," even though they can
both be boundaries of a single colon wall.
[0053] Although the technologies described herein can be used to
identify the entire colon wall for an entire colon, they can also
be used to identify a colon wall for any portion of the colon
(e.g., less than the entire colon, at a point on the colon, or the
like).
[0054] An indication of a boundary of the colon wall can take a
variety of forms. For example, the location and extent of the
boundary can be indicated (e.g., as a set of pixels, voxels, or the
like). The boundary can be of subvoxel accuracy, so the location
can be indicated as a boundary not necessarily limited to discrete
voxels. In practice, the boundary can be represented as a surface.
For example, the outer boundary can be a surface, and the inner
boundary can be a surface.
[0055] Segmentation of the colon wall can be accomplished by
segmenting the inner colon wall boundary and the outer colon wall
boundary. Given both boundaries, the location and extent of the
colon wall can be determined as the space between the two
boundaries. Two-dimensional (e.g., for image slices) or
three-dimensional determinations (for a three-dimensional
representation) can be made.
[0056] The outer boundary and inner boundary can also be considered
together to calculate colon wall thickness. Again, thickness can be
calculated for the entire colon or any portion of the colon (e.g.,
less than the entire colon, at a point on the colon, or the
like).
EXAMPLE 3
Exemplary Subvoxel Accuracy
[0057] In any of the examples herein, subvoxel accuracy can be
achieved. For example, a location for a wall boundary can be a
point in three dimensional space that can refer to fractional
voxels. Similarly, when a wall boundary is expressed as a surface,
the surface can be indicated in units smaller than a single voxel
(e.g., fractional voxels).
EXAMPLE 4
Exemplary Method of Processing a Digital Representation for a Colon
Via a Level Set Technique
[0058] FIG. 2 is a flowchart of an exemplary method 200 of
processing a digital representation for a colon and outputting an
indication of a colon outer wall boundary via a level set technique
and can be implemented, for example, in a system such as that shown
in FIG. 1.
[0059] At 210, a digital representation for a colon is received.
For example, any of the digital representations described herein
can be used.
[0060] At 220, the digital representation is processed via a level
set technique. For example, the colon outer wall boundary can be
identified, segmented, or the like.
[0061] At 230, an indication of the colon outer wall boundary is
outputted. For example, the indication of the colon outer wall
boundary can be outputted to a user for observation (e.g., as a
graphical representation). Or, the indication can be provided to
other computer processing, which can further use the information
(e.g., to calculate wall thickness and the like).
EXAMPLE 5
Exemplary Indication of Colon Wall Outer Boundary
[0062] In any of the examples herein, an indication of the colon
wall outer boundary can take different forms. The indication of the
colon wall outer boundary can take the form of a level set image.
Or, the level set image can be used as an intermediate result from
which a surface (e.g., an isocountour in the level set image
representing the outer colon wall) is provided as an indication of
the colon wall outer boundary. The indication can comprise a
graphical representation (e.g., a displayed representation of) of
the colon wall.
EXAMPLE 6
Exemplary Level Set Processing
[0063] In any of the examples herein, a level set technique can be
used to identify, segment, or otherwise process an outer boundary
for a colon wall. The level set technique can include level set
processing. Level set techniques can evolve an isosurface in the
direction of the surface normal. Evolution speed can depend on
position, normal direction, curvature, shape, and the like. The
isosurface can cross over the same point multiple times.
[0064] A general form of a level set method equation is as
follows:
t .psi. = - .alpha. A .fwdarw. ( x ) .gradient. .psi. - .beta. P (
x ) .gradient. .psi. + .gamma. Z ( x ) .kappa. .gradient. .psi. ( 1
) ##EQU00001##
where .psi. is the level set function, A is an advection term, P is
a speed (e.g., propagation) term, Z is a spatial modifier term for
the mean curvature term .kappa., and .alpha., .beta., and .gamma.
are weights which can determine the influence of the terms on the
movement of the isosurface.
[0065] Another form is as follows:
t .psi. ( x , t ) = - .alpha. A .fwdarw. ( x ) .gradient. .psi. ( x
, t ) - .beta. P ( x ) .gradient. .psi. ( x , t ) + .gamma. Z ( x )
.kappa. .gradient. .psi. ( x , t ) ( 2 ) ##EQU00002##
where .psi.(x, t) is the level set function at point x and time t,
A is an advection term, P is a speed (e.g., propagation) term, Z is
a spatial modifier term for the mean curvature term .kappa., and
.alpha., .beta., and .gamma. are weights which can determine the
influence of the terms on the movement of the isosurface.
[0066] Level set techniques can be used to segment objects in the
presence of noise and incomplete information, with the result
defining the object boundary at subvoxel accuracy. The method can
result in an image that contains positive level set values within
the object and negative level set values external to the object
from which the colon boundaries can be interpolated.
[0067] A general level set method can be applied during
segmentation of both the inner and outer walls from CT virtual
colonoscopy scans. Due to the contrast difference between the colon
wall and lumen, a threshold level set segmentation method (e.g.,
based only on a lower and upper threshold value for the propagation
term) can be used for a subvoxel accurate lumen segmentation.
[0068] To determine the colon outer wall boundary, a more complex
geodesic active contour level set method (e.g., having an advection
term that attracts the level set to the object's boundary) can be
used.
EXAMPLE 7
Exemplary System Outputting an Indication of a Colon Outer Wall
Boundary Via a Speed Image
[0069] FIG. 3 is a block diagram of an exemplary system 300
configured to process a digital representation for a colon and
output an indication of a colon outer wall boundary via a speed
image.
[0070] In the example, a digital representation 312 for a colon is
input for a speed image generator 322, which is configured to
generate a speed image 332 using any of the techniques described
herein.
[0071] The speed image 332 can be used by the level set processor
342 as a speed function during level set processing to output an
indication 382 of the colon outer wall boundary using any of the
techniques described herein.
EXAMPLE 8
Exemplary Speed Function
[0072] In any of the examples herein, a speed image can be used for
a speed function (e.g., a speed term) during level set processing
(e.g., to segment a colon wall, its boundary, or both). The speed
image can correspond to the original digital representation (e.g.,
image) for the colon (e.g., in size, number of components, and the
like). Thus, it can be a three-dimensional image corresponding to
an original acquired three-dimensional CT image, or it can be
derived from the original image using any of a variety of image
processing methods.
[0073] The values for the speed image can be calculated as
described herein to influence the evolution of the level sets on
the original image. Thus, the speed function term value for a
location in the colon at a voxel can be the corresponding value
(e.g., intensity) for the voxel in the speed image.
[0074] For example, intensity values in the speed image can be
calculated as values that encourage or discourage the advance of
the level set (e.g., an isosurface). Thus, during segmentation via
level set processing with a speed function, the speed image can
control the location in the original image at which boundaries are
identified. For example, high intensity values can discourage
evolution of the isosurface, and lower values can encourage
isosurface evolution.
EXAMPLE 9
Exemplary Method of Processing a Digital Representation for a Colon
Via a Level Set Technique with a Speed Image
[0075] FIG. 4 is a flowchart of an exemplary method 400 of
processing a digital representation for a colon via a level set
technique with a speed image and outputting an indication of a
colon outer wall boundary.
[0076] At 410 a digital representation for a colon is received.
[0077] At 440, a speed image for the digital representation is
generated.
[0078] At 450, the colon wall outer boundary is segmented via a
level set technique (e.g., any of the level set processing
described herein). Evolution speed of the isosurface can be proceed
as indicated in the speed image.
[0079] At 460, an indication of the colon outer wall boundary is
outputted. For example, an isocontour in the resulting level set
image can represent the outer wall boundary of the colon.
EXAMPLE 10
Exemplary System Outputting Colon Wall Thickness
[0080] FIG. 5 is a block diagram of an exemplary system 500
configured to process a digital representation for a colon and
output an indication of colon wall thickness therefrom.
[0081] A representation processor (e.g., colon wall segmentation
tool) 522 is configured to receive a digital representation 512 for
a colon as input. The representation processor 522 determines an
indication 526 of the colon wall inner boundary and an indication
528 of the colon wall outer boundary (e.g., via any of the
techniques described herein).
[0082] The representation processor 522 then outputs an indication
of the colon wall thickness 532. For example, the distance between
the boundaries can be calculated. The outer boundary of the colon
wall and the inner boundary of the colon wall can be represented as
surfaces. Distance between the surfaces can be determined. For
example, the minimum distance between points on the inner colon
surface and points on the outer colon surface can be found. For
example, for a point on one (e.g., inner) wall surface (e.g., taken
one at a time from points on the colon surface), the closest point
on the other (e.g., outer) wall surface can be found. The thickness
at the point on the wall surface is the distance between the two
points. Such a technique can be repeated for other (e.g.,
remaining) points on the wall surface. Thickness can be expressed
in millimeters or some other metric.
EXAMPLE 11
Exemplary Method of Outputting Colon Wall Thickness
[0083] FIG. 6 is a flowchart of an exemplary method 600 of
processing a digital representation for a colon and outputting an
indication of colon wall thickness therefrom.
[0084] At 620, a digital representation for a colon is
received.
[0085] At 630, the colon wall inner boundary is segmented (e.g.,
via any of the techniques described herein).
[0086] At 640, the colon wall outer boundary is segmented (e.g.,
via any of the level set techniques described herein).
[0087] At 650, the colon wall thickness is calculated. For example,
colon wall thickness can be calculated (e.g., measured for the
virtual colon) at a particular point (e.g., the site of a polyp
candidate), or at a plurality points. Measurements can be combined
into a single value if desired (e.g., via averaging, median, or the
like).
[0088] At 660, an indication of the colon wall thickness is
outputted. For example, the thickness can be outputted for
processing by software (e.g., a polyp candidate classifier) or
outputted to a user interface for processing by a human user who
can evaluate the thickness. The indication can be in size units
(e.g., millimeters, tenths of millimeters, or the like).
EXAMPLE 12
Exemplary System Generating a Speed Image
[0089] FIG. 7 is a block diagram of an exemplary system 700
configured to process a digital representation for a colon and an
inner wall boundary segmentation to generate a speed image.
[0090] In the example, a digital representation 712 for a colon and
an inner wall boundary segmentation 714 are received by a speed
image generator 722, which generates a speed image 732, which can
be used in any of the examples herein.
EXAMPLE 13
Exemplary Method of Generating a Speed Image
[0091] FIG. 8 is a flowchart of an exemplary method 800 of
generating a speed image from a digital representation for a colon
and an inner wall boundary segmentation and can be used in any of
the examples herein.
[0092] At 810, a digital representation for a colon is
received.
[0093] At 840, the colon wall inner boundary is segmented (e.g.,
via any of the techniques described herein).
[0094] At 860, a speed image is generated via the inner boundary
segmentation and the digital representation for the colon (e.g.,
via any of the techniques described herein). For example, a
derivative of intensity values for an image along a
three-dimensional vector in the direction perpendicular to the
colon wall inner boundary can be calculated. A sigmoid filter can
be applied to the directional derivative image to invert the image
and emphasize values for which the directional derivative is high.
The value can be saved as an intensity value for the speed image at
a location in the speed image corresponding to the location at
which the derivative in the original image was calculated.
[0095] The resulting speed image can be used in any of the examples
herein (e.g., when performing level set processing).
EXAMPLE 14
Exemplary Sigmoid Filter Techniques
[0096] In any of the examples herein, when generating a speed
image, a sigmoid filter can be employed to invert the speed image
after performing directional derivative computations. The filter
can also suppress noise and emphasize high gradient values. Such
high gradient values can reflect the outer wall boundary.
Emphasizing the high gradient values can thus cause the level set
to stop at the outer wall location (e.g., the inverted gradient
value becomes zero). In practice, the sigmoid filter can set the
speed image values to a base level except for the high gradient
values.
EXAMPLE 15
Exemplary System Generating Colon Wall Outer Boundary
[0097] FIG. 9 is a block diagram of an exemplary system 900
configured to process a digital representation for a colon and
generate an indication of a colon wall outer boundary and can be
used in any of the examples herein.
[0098] In the example, a speed image generator 922 is configured to
accept input as a digital representation 902 for a colon and an
inner wall boundary segmentation 914, which is generated by the
inner wall segmenter 912.
[0099] The speed image generator 922 is configured to generate a
speed image 932 based at least on the digital representation 902
and the segmentation 914.
[0100] The outer wall segmenter 942 is configured to generate an
indication 952 of a colon wall outer boundary based at least on the
speed image 932 and the inner wall boundary segmentation 914.
EXAMPLE 16
Exemplary Method of Generating Colon Wall Outer Boundary
[0101] FIG. 10 is a flowchart of an exemplary method 1000 of
generating an indication of a colon wall boundary via a speed image
and a colon inner wall boundary segmentation and can be used in any
of the examples herein.
[0102] At 1010, a digital representation for a colon is
received.
[0103] At 1020, the colon wall inner boundary is segmented (e.g.,
via any of the techniques described herein).
[0104] At 1060, a speed image is generated via the inner boundary
segmentation and a digital representation for the colon.
[0105] At 1070, an indication of the colon wall outer boundary is
generated based at least on the speed image and the colon inner
wall boundary segmentation via level set processing.
EXAMPLE 17
Exemplary Colon Wall
[0106] FIG. 11 is a two-dimensional CT image slice 1100 for a colon
showing colon wall. The example includes various portions that have
a colon wall. For example, a portion 1110 of the image slice
includes a colon wall as described in detail in FIG. 12.
[0107] FIG. 12 is a detail 1200 of FIG. 11 showing the colon outer
wall 1210 and low contrast between the colon outer wall 1210 and
the surrounding tissue. The outer colon wall (e.g., serosal layer)
1210 and inner colon wall (e.g., mucosal layer) 1220 are indicated
on the detail 1200.
EXAMPLE 18
Exemplary Lumen Segmentation
[0108] In any of the examples herein, the lumen can be segmented.
The lumen boundary can be used as (e.g., be the same as) the colon
inner wall. Thus, lumen segmentation can be used as the colon wall
inner boundary segmentation.
[0109] The inner boundary can be useful when segmenting the colon
wall. For example, lumen segmentation can be used as a starting
point for segmenting the colon outer wall. The inner wall can also
be used in conjunction with the outer wall to determine wall
thickness.
[0110] The lumen can be segmented in a variety of ways. For
example, a level set technique can be used that allows for
segmentation of both fluid and air filled regions of the colon.
Such a technique is described in Franaszek et al., U.S. patent
application Ser. No. 11/482,682, filed Jul. 6, 2006, which is
hereby incorporated by reference herein.
[0111] The lumen can be segmented via segmentation that creates a
colon surface. Lumen segmentation can be performed using a simple
threshold region growing method; the large difference in CT
attenuation values between air and colon wall tissue allows the use
of threshold methods to distinguish between the two regions during
the segmentation.
[0112] Another technique is to segment the lumen by combining
threshold region growing with level set methods to result in a
smooth subvoxel-accurate segmentation.
EXAMPLE 19
Exemplary Lumen Segmentation
Region Growing
[0113] A simple threshold region growing segmentation can use a
threshold value (e.g., -500 HU, about -500 HU, or the like) as the
segmentation threshold for the lumen-colon inner wall boundary
because it is the value which is half-way between air (i.e., -1000
HU) and soft tissue (i.e., about 0 HU). Such a segmentation results
in a course lumen segmentation.
EXAMPLE 20
Exemplary Lumen Segmentation
Level Set
[0114] A threshold level set technique for segmenting the lumen can
use the threshold region growing segmentation as an initial level
set boundary and a threshold value (e.g., -500 HU, about -500 HU,
or the like) to determine a subvoxel-accurate segmentation of the
colon lumen. FIG. 13 is a two-dimensional CT image slice 1300
showing superimposed results of lumen segmentation for an image
(e.g., the image slice 1100 of FIG. 11) using such a technique.
[0115] In the example, the lumen (e.g., the colon wall inner
boundary) segmentation results are shown as bright lines, such as
the circular-like line 1310. In the example, subvoxel-accurate
segmentation of the colon lumen was performed.
EXAMPLE 21
Exemplary Speed Function
[0116] In any of the examples herein, a speed function for the
level set technique can be used during segmentation. A speed
function can be represented by a speed image that the level set
segmentation techniques use to determine whether the level set
surface is to evolve and where it is to halt.
[0117] The speed image used in the outer wall level set
segmentation can be calculated from both the lumen level set image
and the original CT image. A three-dimensional directional
derivative of the CT image can be performed in the direction
perpendicular to the level sets produced by the lumen
segmentation.
[0118] FIG. 14 is a detail 1400 of a two-dimensional CT image slice
showing level set isocontours from the inner wall level set
segmentation superimposed on the CT image values. The derivative of
the image calculated in a direction perpendicular to the lumen
level sets can be sigmoid inverted and used as a speed image for
outer wall level set segmentation. The lumen boundary 1410A can be
used as a starting point, from which the speed image is calculated
as a three-dimensional derivative of the CT image in a direction
perpendicular to the level set isocontours 1410A-F produced by the
lumen segmentation. For example, a directed ray (e.g., vector) 1420
is drawn perpendicular to the level set isocontour 1410A in the
example.
[0119] FIG. 15 is a graph showing an intensity profile 1500 of a CT
image and gradient magnitude along a directed ray from the colon
lumen to the outer wall. In the example, the intensity profile of
CT 1510 attenuation values (solid line) and gradient magnitude 1520
(dotted line) values along a directed ray (e.g., ray 1420 of FIG.
14) from the colon lumen to the other wall is shown.
[0120] The local non-maximum gradients along the level set
expansion direction can be suppressed to further avoid the impact
from noise and partial voluming effect from the lumen-colon wall
boundary. A sigmoid filter can be used on the directional
derivative image emphasizing the particular set of values where the
directional derivative is high (e.g., where the outer colon wall
boundary is located).
[0121] Inverting the output of the sigmoid filter allows a speed
image to be created such that the level sets will propagate where
there is a low directional gradient in the original CT image and
stop when a high gradient along the colon outer wall is
encountered.
EXAMPLE 22
Exemplary Colon Outer Wall Segmentation
[0122] In any of the examples herein, the level set segmentation of
the colon outer wall can be computed via a three-dimensional
geodesic active contour level set segmentation technique. The lumen
level set segmentation can be used as the initial level set
boundary, and the speed image can be calculated from the
directional derivative of the original CT image (e.g., as described
in Example 21).
[0123] The geodesic active contour level set segmentation technique
can use an advection term that attracts the level set evolution to
the high gradient values in the feature image and a curvature term
that prevents the evolution of the boundary from exceeding a
maximum curvature. The level set technique can adhere to near zero
values (e.g., all near zero values) in the speed image and fill in
the missing regions as desired, producing an outer wall
segmentation that combines the confident location of boundaries
seamlessly with desired boundaries. An isocontour (e.g., zero, two,
or the like) in the resulting level set image can be used to
represent the outer colon wall.
EXAMPLE 23
Exemplary Results
[0124] The technologies described herein were performed on three CT
virtual colonoscopy scans each containing 512.times.512.times.512
images with a spacing of 0.7.times.0.7.times.1.0 mm.sup.3. The
colon wall in the scans consisted of various thicknesses throughout
each colon segment. The results of performing the segmentation
technologies on these cases is shown in FIG. 16-20;
subvoxel-accurate segmentation of the colon wall was performed.
[0125] FIG. 16 shows a graphical representation 1600 of a colon
with a cutting plane showing segmentation of the lumen (green) and
outer colon wall (purple) as determined via the technologies
described herein. A detail 1610 shows a close up view of the colon
wall.
[0126] FIGS. 17, 18, and 19 show two-dimensional CT image slices
1700, 1800, and 1900 showing superimposed results of colon wall
segmentation performed according to the technologies described
herein. The colon inner wall boundary (e.g., as determined via the
techniques described herein) is shown in green, and the colon outer
wall (e.g., as determined via the techniques described herein)
boundary is shown in red.
[0127] The accuracy of the segmentations was verified visually. The
outer boundary of the colon wall was determined accurately, even
though there was low contrast between the colon wall and the
surrounding fat tissue. The technique can be fully automatic, thus
requiring no user intervention.
[0128] Even in areas where the colon is adjacent to other organs,
the technologies can accurately find the colon outer wall, as shown
in FIG. 20, which is a detail 2000 of a two-dimensional CT image
slice showing superimposed results of colon wall segmentation
performed according to the technologies described herein. The colon
inner wall boundary is shown in green, and the colon outer wall
boundary is shown in red. The detail 2000 shows an area of colon
wall segmentation adjacent to small bowel in the upper portion of
the detail 2000.
[0129] Using the derivative of the CT values along the direction
perpendicular to the level set surfaces of the lumen segmentation
allowed for an accurate detection of the colon outer wall. Further,
the use of the geodesic level set method has allowed for a smooth
subvoxel accurate colon wall segmentation to be performed.
[0130] When determining the position of the colon outer wall by
starting at the lumen segmentation and considering the gradient
direction relative to the lumen level set gradients, partial
voluming effects between the colon lumen and wall can be avoided.
Partial voluming effects cause difficulty in accurately segmenting
the colon wall.
[0131] Also, finding the outer boundary within the lumen can be
avoided if the technique is initialized with the lumen
segmentation.
[0132] The resulting segmentation can contain the entire surface,
rather than only several points on the outer colon boundary.
EXAMPLE 24
Exemplary Enhanced Level Set Techniques
[0133] In any of the examples herein, level set segmentation of the
colon outer wall can be computed by using a three-dimensional
geodesic active contour level set segmentation method.
[0134] A lumen level set segmentation can be used as the initial
level set boundary, and the speed image can be calculated from a
directional derivative of the original CT image. A
three-dimensional derivative of the CT image can be performed in a
direction perpendicular to the level sets produced by the lumen
segmentation:
g ( ) = k ( + v ( ) ) - k ( - v ( ) ) 2 ( 3 ) ##EQU00003##
where k(x) is the CT value at position x, and v(x) is the vector
(e.g., the vector 1420 of FIG. 14) perpendicular to the
segmentation level sets at position x.
[0135] The local non-maximum gradients along the level set
expansion direction can be suppressed to avoid the impact from
noise and partial voluming effects from the lumen-colon wall
boundary by removing isolated pixels of high gradient
magnitudes.
[0136] By using two sigmoid filters in series on the directional
derivative image with an .alpha.=-4.0, a .beta.=0.02 and a
.beta.=0.0, .beta.=0.48, respectively, both with min=0 and max=1,
the particular set of values where the directional derivative is
high (i.e., where the outer colon wall boundary is located) can be
emphasized. Inverting the output of the sigmoid filter allows a
speed image to be created such that the level sets will propagate
where there is a low directional gradient in the original CT image
and stop when a high gradient along the colon outer wall is
encountered.
[0137] FIG. 21 shows an exemplary speed image 2100. The speed image
can be used to determine the propagation of the geodesic active
contour level set segmentation of the colon outer wall. White
indicates high speeds of propagation, while black indicates zero
speed of propagation. The colon inner boundary is shown in
green.
[0138] The geodesic active contour level set segmentation can then
be used along with the speed image to determine the location of the
colon outer wall. This method can use Equation (2) with an
advection term, .alpha.=0.3, that attracts the level set evolution
to the high gradient values in the feature image, a propagation
term, .beta.=0.2, that evolves the boundary outwards, and a
curvature term, .gamma.=0.3, that prevents the evolution of the
boundary from exceeding a maximum curvature. The level set
technique also adheres to near zero values (e.g., all near zero
values) in the speed image and fills in the missing regions to
produce a boundary as desired. The geodesic active contour level
set segmentation method produces an outer wall segmentation that
combines the confident location of boundaries seamlessly with the
expected boundaries. The zero isosurface in the resulting level set
image can represent the outer colon wall.
[0139] FIG. 22 is a two-dimensional CT image slice 2200 showing
superimposed results of colon wall segmentation via the speed image
shown in FIG. 21. The colon inner boundary (e.g., as determined via
the techniques described herein) is shown in green, and the colon
outer boundary (e.g., as determined via the techniques described
herein) is shown in red.
EXAMPLE 25
Exemplary Outer Wall Segmentation
[0140] FIG. 23 is an illustration of a surface 2300 of an outer
colon wall determined via level set segmentation (e.g., using the
inner wall segmentation as an initial surface for the colon's outer
wall segmentation).
EXAMPLE 26
Exemplary Uses of Outer Wall Segmentation
[0141] In any of the examples herein, the outer wall segmentation
can be useful for segmenting the colon wall and automatically
determining its position, which can otherwise be difficult (e.g.,
because of the low contrast between CT attenuation values for the
colon wall and the surrounding fat tissue). For example, colon wall
thickness can be calculated using the outer wall segmentation in
combination with the inner wall segmentation. So, colon wall
thickness can be calculated automatically by software.
[0142] The wall thickness thus determined can have desirable
properties, such as relatively smooth variation and decreased
sensitivity to noise.
[0143] Further, segmenting the outer wall can be useful for
identifying and classifying polyp candidates. For example,
characteristics of the outer wall, colon wall thickness, or both
can be included as an input feature when identifying or classifying
polyp candidates.
[0144] Another use of the outer wall segmentation is identifying
colonic diseases (e.g., muscular hypertrophy and diverticulitis),
which can be diagnosed via colon wall thickness calculation
results.
[0145] Still other uses are colon spasm detection, and colon cancer
detection.
[0146] Still other uses include determining a colon centerline. The
outer colon wall or the centerline can be used for determining a
path for a fly through of the virtual colon (e.g., in portions of
the colon which are insufficiently distended to allow for
segmentation of the colonic inner wall).
[0147] By contrast, manual determination of the colon wall
thickness can be very time consuming. In addition, it is often
difficult to determine the precise location of the boundaries of
the colon on two-dimensional slices due to partial voluming
effects.
EXAMPLE 27
Exemplary Polyp Candidate Classifier
[0148] In any of the examples described herein, colon wall
thickness can be inputted into a polyp candidate classifier
configured to determine whether a polyp candidate is a true
positive. For example, a software system can identify polyp
candidates. A polyp candidate classifier can then classify the
polyp candidates as true positives or false positives based on a
plurality of features (e.g., characteristics) submitted to the
polyp candidate classifier. The classifier can use the colon wall
thickness calculations (e.g., based on level set segmentation of
the colon wall) described herein as an input feature.
[0149] Such classifiers can be trained or otherwise developed based
on training data with known results (e.g., polyp candidates
classified by a human radiologist).
EXAMPLE 28
Exemplary Automatic Technologies
[0150] The technologies described herein can provide an automatic
technique for determining a subvoxel-accurate segmentation of the
colon wall. The technologies can result in appropriate
segmentations when the colon wall is thick or thin. The resulting
accurate segmentations of the colon wall can be useful for any of
the applications described herein.
EXAMPLE 29
Exemplary User Interfaces
[0151] In any of the examples herein, graphical depiction of the
colon wall or a portion thereof can be displayed to a human (e.g.,
radiologist), who decides what action, if any, to take. Such
interfaces can allow manipulation of the graphical depiction, such
as rotation, zooming, and the like.
[0152] The interface can highlight (e.g., zoom in on or depict in a
special color) areas detected as unusual (e.g., having colon wall
thickness outside of defined thresholds).
EXAMPLE 30
Exemplary Colon Thickness Calculation
[0153] After corresponding points on both the inner and outer
surface are calculated, the Euclidean distance between the outer
and inner surface at each potential polyp position can be
calculated.
EXAMPLE 31
Exemplary Level Set Processing
[0154] A Laplacian level set method can be used to perform the
lumen segmentation and results in a level set image that contains
different level set isosurface values; the zero isosurface
represents the lumen-colon wall boundary. Since the outer wall
segmentation can use the lumen segmentation as the initial surface,
the value of the level set function in the lumen segmentation image
at any point is the distance from the point to the current front.
Thus, for any point on the outer surface, the absolute value of the
lumen segmentation level set field at the point is the level set
distance from the point to the inner surface. A threshold on the
distance can be used to remove areas where the colon wall thickness
is low. The threshold can also simultaneously eliminate many folds
from a list of potential polyps.
EXAMPLE 32
Exemplary Colon Thickness Map
[0155] In any of the examples herein, a thickness map between the
inner and outer colon wall surfaces can be assembled from the
calculated thicknesses. The map can be depicted visually by showing
a graphic representation of the colon and using different colors to
represent different thicknesses (e.g., one color for average
thickness, another for above average, and another for below
average). The colon wall thickness can be color mapped on the colon
surface.
[0156] If desired, opacity for the average and below average colors
can be varied to make the regions of interest (e.g., above average
thickness) more visible.
EXAMPLE 33
Exemplary Clustering and Filtering
[0157] In any of the examples herein, after a thickness map of the
colon has been calculated, the list of potential polyps can be
further reduced by clustering candidate detections that are close
to each other (e.g., within a threshold distance). For example,
potential polyp voxels that are within n (e.g., two) voxels from
each other can be grouped into the same polyp location. Also,
detected points that have only n (e.g., one) voxels can be
eliminated because such detections are most likely due to
noise.
[0158] One technique for determining where a polyp is located is to
use a threshold thickness. The thickness can also be used as one of
a plurality of features of a more sophisticated classifier, such as
a support vector machine or one or more neural networks. The
classifier can be trained on known polyps.
EXAMPLE 34
Exemplary Colon Thickness Map Experiments
[0159] The level set techniques described herein for initial polyp
detection was compared to a colonography CAD system that uses mean
curvature, Gaussian curvature, and sphericity to detect potential
polyps on the colon surface. The curvature and sphericity threshold
parameter settings, which allow for the detection of elliptical
shaped objects, were set to predetermined values. Polyp detections
using the curvature based method that had only one point were
eliminated from the list of potential polyp detections.
[0160] The technologies described herein were performed on three
randomly chosen CT virtual colonoscopy scans with volume sizes
between 512.times.512.times.354 to 512.times.512.times.424 images
with a spacing of 0.7.times.0.7.times.1.0 mm.sup.3. Each colon
contained one polyp, which had a size of 1.5 cm, 2.0 cm, and 1.0
cm, respectively. The colon wall in the scans had various
thicknesses throughout the colon segments.
[0161] The results of performing level set based segmentation and
thickness calculation on the three cases were visualized as shown
in FIGS. 24, 25, 26 and 27A-C. FIGS. 24, 25, and 26 show the
results of color mapping the thickness on the colon wall. Thicker
colon wall areas are indicated by red, while thinner colonic areas
are indicated by blue, and average thickness for the particular
colon is indicated by the color green.
[0162] The average colonic wall thickness was computed to be
4.02.+-.1.80 mm, 4.91.+-.2.04 mm, and 3.66.+-.1.83 mm,
respectively. The colonic wall thickness at the polyp location in
the colons was 8.0 mm, 10.0 mm, and 4.0 mm, respectively. Regions
of interest (e.g., high wall thickness), indicated by the red
regions are further visible by lowering the opacity of the blue and
green areas, as shown in FIGS. 27A-C.
[0163] FIGS. 28, 29, and 30 show the results using the wall
thickness to detect potential polyp candidates. The colon inner
wall is shown in green; the outer wall is shown in red. Polyp
detections are indicated in purple, while false positive detections
(e.g., polyp candidates identified but actually not polyps) are
shown in blue. The computed results were compared to the optical
colonoscopy-proven polyp locations to determine the accuracy of the
method. Each polyp was detected by both the wall thickness based
and curvature based methods. Table 1 shows the number of false
positives that resulted from both of these methods in the analysis
of each colon.
TABLE-US-00001 TABLE 1 The number of potential polyp candidates
calculated based on the curvature and wall thickness methods for
the three different colons shown herein. Number of Number of
Potential Polyps Potential Polyps Percentage Based on Based on Wall
Potential Polyps Colon Curvature Thickness Reduction 1 1,163 200
82.8% 2 976 499 48.9% 3 1,274 708 44.4%
EXAMPLE 35
Exemplary Further Information
[0164] Within the three colons, the majority of the segments have
areas that have an average thickness for the respective colon with
only isolated areas of increased colon wall thickness. When
comparing the computed results to the ground truth (e.g., optical
colonoscopy), the polyps in the colons are located in areas that
have high colon wall thickness; these results are shown in the
zoomed in areas of FIGS. 27A-C. The thick areas of the colon which
were not present in the ground truth data as polyps are due to
normal variation of the colon wall thickness throughout the colon,
due to haustral folds in the colon, due to the different amounts of
distention in various areas of the colon, and due to lack of a
detectible edge along the outer wall.
[0165] The polyps in the three scans were detected by the wall
thickness method. Table 1 indicates that the wall thickness method
for initial polyp detection results in a reduced number of false
positives compared to the curvature based method performed on the
same colon. In the example, the techniques involved only initial
detection of polyps without feature extraction for each polyp and
classification to further eliminate false positives. The majority
of the false positives detected by the wall thickness method
consisted of enlarged colon wall thickness due to folds and due to
difficulty in the lumen segmentation near air-fluid boundaries.
EXAMPLE 36
Exemplary Wall Thickness Techniques
[0166] In any of the examples herein, the thickness of the colon
wall throughout the colon can be determined. However, such a
computation can be time consuming. A binary space partitioning tree
can be used to speed up the calculation of the minimum distance
between two surfaces at points. Using the partitioning tree, the
calculation can be performed in O(n log(m)) time, where n is the
number of points on the outer surface, and m is the number of
points on the lumen surface.
EXAMPLE 37
Exemplary Level Set Based Thickness Technique
[0167] The level set methods described herein for determining the
location of the colon outer wall and calculating colonic wall
thickness can be used for visually assessing the thickness
variations across different regions of the colon and for the
initial detection of polyps in the colon. In the experiments, all
polyps were detected by the wall thickness method, and the number
of false positives generated for each colon was between 44.4% and
82.8% less than the curvature based method. The wall thickness
calculation can be used in conjunction with the curvature based
method for the detection of polyps to further reduce the final
number of false positives detected.
EXAMPLE 38
Exemplary Detection of Colonic Diverticular Disease
[0168] The technologies described herein can be used to detect
colonic diverticular disease. Estimates state that one third of all
individuals have some form of colonic diverticular disease by the
age of 50, and approximately two-thirds are affected by the age of
80. Colon wall thickness is a property characteristic of
diverticular disease. Accordingly, the techniques described herein
can be applied to determine colon wall thickness and detect colonic
diverticular disease therefrom.
EXAMPLE 39
Exemplary Techniques for Detection of Colonic Diverticular
Disease
[0169] The inner and outer walls of the colon can be segmented, and
the thickness of the colon wall can be determined. While a normal
colonic wall will have a thickness of 1-4 mm, a colonic wall where
diverticular disease is present could have a thickness of as much
as 10 mm or more. When colonic diverticular disease is present, the
thickness of the wall is several millimeters thicker in affected
segments (e.g., areas) of the colon than in normal segments of the
colon. FIG. 31 is a slice 3100 of a CTC scan showing colonic wall
affected by diverticular disease. The portion 3110 of the slice
having thickened colonic wall is shown in FIG. 32 as an inset 3200.
The outer colon wall (serosal layer) 3210 and the inner colon wall
(mucosal layer) 3220 are shown in the inset 3200.
[0170] In practice, a variety of techniques can be used to detect
diverticular disease. A method for detecting diverticular disease
can be to determine wall thickness (e.g., via any of the techniques
described herein) and then applying any of a variety of techniques
alone or in combination. For example, a threshold thickness can be
used. A range of intensities of the colon wall can be used.
Detections can be clustered, and various features can be computed
per cluster. A support-vector machine can be used to classify
whether the features (e.g., for a point, location, or cluster)
indicate diverticular disease of the colon.
EXAMPLE 40
Exemplary Segmentation
[0171] FIG. 33 is a slice of a CTC scan showing segmentation of the
colon wall as performed via the techniques described herein (e.g.,
using a value of -500 HU for the lumen-colon inner wall boundary
segmentation threshold and determining a subvoxel precise
segmentation). The colon inner boundary is shown in green, and the
colon outer boundary is shown in red.
EXAMPLE 41
Exemplary Filters for Detection of Diverticular Disease
[0172] In any of the examples herein, linear filters can be used to
eliminate areas of the colonic outer wall where diverticular
disease is not present. Initially, an entire colon or area (e.g.,
sigmoid colon) thereof can be considered candidate detections.
Filters can remove areas that are considered to not have
diverticular disease. The filters can be used in conjunction to
reduce false positives that remain. For example, a thickness filter
and an intensity filter can be applied to reduce false
positives.
[0173] For example, a filter can be a threshold based on the
thickness of the colon wall at the position. A colon wall thickness
threshold (e.g., 4 mm or the like) can be used as a threshold for
determining the possible presence of diverticular disease. If the
wall thickness is less than the threshold, the location or point on
the colon can be classified as non-diverticular (e.g., normal).
False positives can remain after using the first filter.
[0174] Another filter can be the intensity of the colon wall. If
the location or point on the outer wall does not contain intensity
values between two thresholds experimentally determined as expected
values for the colonic wall, the location or point can be
classified as being non-diverticular (e.g., diverticular disease is
not detected because the colon wall is not involved). For example,
a range between -50 and 550 HU or the like can be used.
EXAMPLE 42
Exemplary Clustering for Detection of Diverticular Disease
[0175] Because diverticular disease is not a localized disease but
will be present throughout whole segments of the colon, detection
candidates within a threshold distance can be clustered into a
single candidate detection. For example, detections within an n
(e.g., 10 or about 10) pixel neighborhood meeting specified
criteria can be clustered together and considered a single
detection.
[0176] For respective clusters (e.g., each cluster), features can
be calculated. Such features can include average and standard
deviation of the colon wall thickness, and average and standard
deviation of the CT intensity values.
[0177] The features can then be considered to determine whether the
clustered detection indicates colonic diverticular disease.
EXAMPLE 43
Exemplary Classification of Features
[0178] In any of the examples herein, a support-vector machine
(SVM) can be used for classifying detected features to determine
whether a candidate detection indeed indicates colonic diverticular
disease. Classification can be performed based on features
calculated for detection clusters (e.g., average and standard
deviation of the colon wall thickness, and the average and standard
deviation of the CT intensity values). Ground truth detection of
diverticular disease (e.g., for purposes of configuring the
classifier) can be determined by visual inspection of the CT images
by a qualified professional. Sensitivity can be calculated on a
per-diverticular disease detection basis, and a free-response
operating characteristic (FROC) curve can be plotted.
EXAMPLE 44
Exemplary Experimental Results
[0179] The diverticular disease detection techniques described
herein were performed on ten (10) CT colonoscopy scans each
containing volume sizes between 512.times.512.times.380 to
512.times.512.times.512 images with a spacing of
0.7.times.0.7.times.1.00 mm.sup.3. Five of the CT scans contained
colonic diverticular disease and the other five were from normal
patients. Results of the detection of diverticular disease in the
sigmoid segment of two different colons can be seen in FIGS. 34 and
35.
[0180] FIGS. 34 and 35 are slices of CTC scans showing the sigmoid
portion of two different colons with diverticular disease. The
colon inner boundary is shown in green, the colon outer boundary is
shown in red, and areas (e.g., clusters) detected as having
diverticular disease are shown as blue squares. Although a
two-dimensional view is shown, in practice, the locations can be
indicated as two-dimensional or three-dimensional locations (e.g.,
via a coordinate system).
[0181] The technique produced 123 detections for the ten data sets,
where 87 of them were true positives, and the remaining 36 were
false positives. Due to the extent of diverticular disease,
multiple true detections may be produced for one disease location.
The technique successfully detected all patients having
diverticular disease, which corresponds to a sensitivity of 100% at
3.6 false positives per patient.
[0182] In order to reduce the false positives further, a
support-vector machine (SVM) classifier was run on the four
calculated features and produced FROC curves for two cases: one
where multiple detections were merged to calculate a sensitivity
per patient, and one where multiple detections were not merged to
calculate a sensitivity per detection. Merging multiple detections
means that any detection for a diverticular disease location was
considered as correctly classifying the diverticular disease. In
the not merging case, each misclassified detection was counted as a
misclassification.
[0183] FIG. 36 is a graph of a free-response receive operating
characteristic curve for four features calculated on diverticular
detection candidates and shows that if the multiple true detections
are merged the false positives can be reduced to 0.2 per patient at
a sensitivity of 100%. If the multiple true detections are not
merged, a sensitivity of 91% can still be obtained for the 87 true
detections at 1.1 false positives per patient. The top curve shows
sensitivity per patient, and the bottom one shows sensitivity per
detection.
EXAMPLE 45
Exemplary Further Information
[0184] The results shown in FIGS. 34 and 35 show the accurate
segmentation of the colon outer wall. The outer colon wall surfaces
generated are subvoxel precise and can be generated fully
automatically.
[0185] The results of the SVM classification demonstrate the
ability of the technique to detect colonic diverticular disease.
The FROC curve shows that the false positives of the technique can
be 0.2 per patient at a sensitivity of 100%, if the multiple true
detections are merged. If the multiple true detections are not
merged, the sensitivity for all true detections can be 91% at 1.1
false positives per patient.
[0186] Many of the false positives detected by the technique are
due to inaccurate segmentation of fluid-filled regions in the colon
lumen. The lack of a visible colon wall in normal, well distended
portions of the colon also contributes to false positive
detections. The other major source of false positives is from the
technique mistakenly identifying as colonic wall other organs that
abut the colon and have similar CT intensity values as the colon
wall, such as the small bowel.
EXAMPLE 46
Exemplary Focus
[0187] In any of the examples herein, detection of diverticular
disease can be performed for the entire colon or limited to one or
more areas therein. For example, detection can be limited to the
sigmoid colon because it is the area of the colon that is most
often affected by colonic diverticular disease. For example, the
digital representation (e.g., image) can be cropped to contain only
the sigmoid colon or non-sigmoid areas can be otherwise omitted
from analysis.
EXAMPLE 47
Exemplary Anatomical Structures
[0188] Although many of the examples herein describe a colon, the
technologies can also be applied to any of the other anatomical
structures described herein.
[0189] The technologies can also be applied to other scenarios
involving two or more concentric cylinders (e.g., tree rings,
pipes, fruit, or the like).
EXAMPLE 48
Exemplary Anomalies of Interest
[0190] Any of the examples herein describing polyp candidates can
be applied to anomalies of interest. Exemplary anomalies of
interest include noncancerous growths, precancerous growths, and
cancerous growths. Such anomalies include polyps, which are growths
associated with mucus membranes and epithelial linings. Polyps of
interest include colonic, small intestine, nasal, urinary tract,
and uterine polyps. Other exemplary anomalies of interest include
atherosclerosis and instances of hyperplasia: an abnormal growth of
the lining of an organ.
[0191] It is important that polyps and other anomalies be detected
because they can be premalignant and if detected can be
prophylactically removed to avoid development of diseases such as
gastrointestinal adenocarcinoma. Thus, early detection enables
early treatment (such as removal of the polyp) of possibly
life-threatening conditions.
[0192] In any of the examples herein, any of the anomalies detected
in a digital representation can be analyzed to detect anomalies of
interest which correspond to anomalies of interest in the
represented real world anatomical structure. Various software
filtering mechanisms as described herein can be used on an initial
list of detected anomalies of interest (e.g., polyp candidates) to
provide a resulting list of anomalies of interest (e.g., confirmed
candidates).
EXAMPLE 49
Exemplary Improvements Gained by Using Exemplary Embodiments
Herein
[0193] The embodiments disclosed herein present a segmentation
technique that can be implemented fully-automatically, and which
does not require user interaction.
EXAMPLE 50
Exemplary Acquisition of Digital Representations
[0194] A variety of technologies can be used to acquire
three-dimensional digital representations for use with the
technologies described herein. Acquisition of a representation of
an anatomical structure is typically done by performing a scan of
the soft tissues of the patient. For example, a CT scan can be
performed according to any number of standard protocols. CT scans
can be used to generate thin-section CT data (for example, helical
scan CT data). The representation can be analyzed immediately after
the scan, or the representation can be stored for later retrieval
and analysis. Exemplary technologies for acquiring scans are
described in Pickhardt et al., "Computed Tomographic Virtual
Colonoscopy to Screen for Colorectal Neoplasia in Asymptomatic
Adults," New Engl. J. Med., 349:2191 (2003), Vining et al.,
"Virtual Colonoscopy," Radiology 193(P):446 (1994), Vining et al.,
"Virtual Bronchoscopy," Radiology 193(P):261 (1994), and Vining et
al., "Virtual bronchoscopy. Relationships of virtual reality
endobronchial simulations to actual bronchoscopic findings" Chest
109(2): 549-553 (February 1996).
[0195] Any number of hardware implementations can be used to
acquire a representation of an anatomical structure. For example,
the GE HiSpeed Advantage scanner of GE Medical Systems, Milwaukee,
Wis. can be used.
[0196] Additional exemplary segmentation technologies are described
in U.S. Pat. No. 6,556,696 to Summers et al., filed Feb. 5, 2002,
entitled, "METHOD FOR SEGMENTING MEDICAL IMAGES AND DETECTING
SURFACE ANOMALIES IN ANATOMICAL STRUCTURES," which is hereby
incorporated herein by reference.
EXAMPLE 51
References
[0197] The following references are hereby incorporated by
reference herein: [0198] [1] A. Jemal, R. C. Tiwari, T. Murray, A.
Ghafoor, A. Samuels, E. Ward, E. J. Feuer, M. J. Thun. Cancer
statistics, 2004. CA Cancer J Clin, 54:8-29, 2004. [0199] [2] T. M.
Gluecker, C. D. Johnson, W. S. Harmsen, K. P. Offord, A. M. Harris,
L. A. Wilson, D. A. Ahlquist. Colorectal cancer screening with CT
colonography, colonoscopy, and double-contrast barium enema
examination: prospective assessment of patient perceptions and
preferences. Radiology, 227:378-84, 2003. [0200] [3] R. M. Summers,
J. Yao, P. J. Pickhardt, M. Franaszek, I. Bitter, D. Brickman, V.
Krishna, J. R. Choi. Computed Tomographic Colonoscopy
Computer-Aided Polyp Detection in a Screening Population.
Gastroenterology, 129: 1832-1844, 2005. [0201] [4] J. A. Sethian.
Level Set Methods and Fast Marching Methods: Evolving Interfaces in
Computational Geometry, Fluid Mechanics, Computer Vision, and
Materials Science. Cambridge University Press, 1999. [0202] [5] R.
Kimmel, V. Caselles, G. Saprio. Geodesic active contours.
International Journal on Computer Vision, 22(1):61-97, 1997. [0203]
[6] R. M. Summers, A. K. Jerebko, M. Franaszek, J. D. Malley, C. D.
Johnson. Colonic Polyps: complementary role of computer-aided
detection in CT colonography. Radiology 225:391-399, 2002. [0204]
[7] M. Franaszek, R. M. Summers, P. J. Pickhardt, J. R. Choi.
Hybrid Segmentation of Colon Filled with Air and Opacified Fluid
for CT colonography. IEEE Tras. Med. 1 mg. 25: 358-368, 2006.
[0205] [8] Z. Wang, A. Liang, L. Li, X. Li, B. Li, J. Anderson, D.
Harrington. Reduction of false positives by internal features for
polyp detection in CT-based virtual colonoscopy. Medical Physics,
32: 3602-3615, 2005. [0206] [9] Z. Zeng, L. H. Staib, R. T.
Schultz, and J. S. Duncan. Segmentation and Measurement of the
Cortex from 3-D MR Images Using Coupled--Surfaces Propagation. IEEE
Trans. Med. 1 mg, 18:927-937, 1999.
EXAMPLE 52
Exemplary Computer System for Conducting Analysis
[0207] FIG. 37 and the following discussion provide a brief,
general description of a suitable computing environment for the
software (for example, computer programs) described above. The
methods described above can be implemented in computer-executable
instructions (for example, organized in program modules). The
program modules can include the routines, programs, objects,
components, and data structures that perform the tasks and
implement the data types for implementing the technologies
described above.
[0208] While FIG. 37 shows a typical configuration of a desktop
computer, the technologies may be implemented in other computer
system configurations, including multiprocessor systems,
microprocessor-based or programmable consumer electronics,
minicomputers, mainframe computers, and the like. The technologies
may also be used in distributed computing environments where tasks
are performed in parallel by processing devices to enhance
performance. For example, tasks related to measuring
characteristics of anomalies of interest can be performed
simultaneously on multiple computers, multiple processors in a
single computer, or both. In a distributed computing environment,
program modules may be located in both local and remote memory
storage devices.
[0209] The computer system shown in FIG. 37 is suitable for
implementing the technologies described herein and includes a
computer 3720, with a processing unit 3721, a system memory 3722,
and a system bus 3723 that interconnects various system components,
including the system memory to the processing unit 3721. The system
bus may comprise any of several types of bus structures including a
memory bus or memory controller, a peripheral bus, and a local bus
using a bus architecture. The system memory includes read only
memory (ROM) 3724 and random access memory (RAM) 3725. A
nonvolatile system (for example, BIOS) can be stored in ROM 3724
and contains the basic routines for transferring information
between elements within the personal computer 3720, such as during
start-up. The personal computer 3720 can further include a hard
disk drive 3727, a magnetic disk drive 3728, for example, to read
from or write to a removable disk 3729, and an optical disk drive
3730, for example, for reading a CD-ROM disk 3731 or to read from
or write to other optical media. The hard disk drive 3727, magnetic
disk drive 3728, and optical disk 3730 are connected to the system
bus 3723 by a hard disk drive interface 3732, a magnetic disk drive
interface 3733, and an optical drive interface 3734, respectively.
The drives and their associated computer-readable media provide
nonvolatile storage of data, data structures, computer-executable
instructions (including program code such as dynamic link libraries
and executable files), and the like for the personal computer 3720.
Although the description of computer-readable media above refers to
a hard disk, a removable magnetic disk, and a CD, it can also
include other types of media that are readable by a computer, such
as magnetic cassettes, flash memory cards, DVDs, and the like.
[0210] A number of program modules may be stored in the drives and
RAM 3725, including an operating system 3735, one or more
application programs 3736, other program modules 3737, and program
data 3738. A user may enter commands and information into the
personal computer 3720 through a keyboard 3740 and pointing device,
such as a mouse 3742. Other input devices (not shown) may include a
microphone, joystick, game pad, satellite dish, scanner, or the
like. These and other input devices are often connected to the
processing unit 3721 through a serial port interface 3746 that is
coupled to the system bus, but may be connected by other
interfaces, such as a parallel port, game port, or a universal
serial bus (USB). A monitor 3747 or other type of display device is
also connected to the system bus 3723 via an interface, such as a
display controller or video adapter 3748. In addition to the
monitor, personal computers typically include other peripheral
output devices (not shown), such as speakers and printers.
[0211] The above computer system is provided merely as an example.
The technologies can be implemented in a wide variety of other
configurations. Further, a wide variety of approaches for
collecting and analyzing data related to processing anomalies of
interest is possible. For example, the data can be collected,
characteristics determined and measured, anomalies classified and
reclassified, and the results presented on different computer
systems as appropriate. In addition, various software aspects can
be implemented in hardware, and vice versa.
EXAMPLE 53
Exemplary Methods
[0212] Any of the methods described herein can be performed by
software executed by software in an automated system (for example,
a computer system). Fully-automatic (for example, without human
intervention) or semi-automatic operation (for example, computer
processing assisted by human intervention) can be supported. User
intervention may be desired in some cases, such as to adjust
parameters or consider results.
[0213] Such software can be stored on one or more computer-readable
media (e.g., storage media or other tangible media) comprising
computer-executable instructions for performing the described
actions (e.g., causing a computer to perform actions of the methods
shown).
ALTERNATIVES
[0214] Having illustrated and described the principles of the
invention in exemplary embodiments, it is noted that the described
examples are illustrative embodiments and can be modified in
arrangement and detail without departing from such principles.
Technologies from any of the examples can be incorporated into one
or more of any of the other examples.
[0215] In view of the many possible embodiments to which the
principles of the invention may be applied, it should be understood
that the illustrative embodiments are intended to teach these
principles and are not intended to be a limitation on the scope of
the invention. We therefore claim as our invention all that comes
within the scope and spirit of the following claims and their
equivalents.
* * * * *